A variety of screening tools identify children who are at risk for autism. Level 1 tools are used in unselected samples, but tend to have high false positive rates. Reducing the number of false positives cases will decrease the delay in receiving intervention services for true positive cases.
Objectives:
This study seeks to measure whether a multilevel screening method, using the STAT (Level 2) following screen positive results on the M-CHAT-R and Follow-up Interview (FUI; Level 1), will reduce false positives without significantly increasing the number of missed cases. Additionally, this study seeks to replicate the utility of the STAT with children younger than 24 months of age.
Methods:
Parents in the metro-Atlanta area completed the M-CHAT-R at their child’s well-baby visits (n=6,914), 640 of whom screened positive; 99 completed the FUI and continued to screen positive. A subsample of 44 children completed both a STAT (Level 2) and a diagnostic evaluation. STAT cutoffs of 2.00 for children ≥ 24 months and 2.75 for children < 24 months were based on recommendations in Stone, McMahon, and Henderson (2008).
Results:
Sixteen cases screened positive on the STAT, of which 13 received an ASD diagnosis. This multilevel screening method yielded a PPV of .81, compared to preliminary analyses using the M-CHAT-R & FUI alone, which suggest a PPV of .59. Half of the 28 children who screened negative on the STAT had ASD, resulting in inadequate sensitivity of .48 using multilevel screening.
ROC analysis for the subsample of children who completed a STAT before 24 months and also received a diagnostic evaluation (n=23) yielded an area under the curve (AUC) of .69. An optimal STAT cutoff score for this age group was 2.25: sensitivity=.80 and specificity=.71. Psychometric properties of the STAT for children ≥ 24 months (n=21) indicated that a cutoff of 2.00 was optimal: sensitivity=.82 and specificity=.80 (AUC=.88).
Reanalysis of the data for two-level screening using STAT cutoffs of 2.25 and 2.00 by age resulted in one additional ASD case being misclassified as low-risk, but correctly re-classified 8 of the 14 false negatives as true positives. These new cutoffs maintained a strong PPV (.84) and increased sensitivity to .78, which was a 63% improvement compared to using published cutoffs.
Conclusions:
The difference in optimal cutoffs in the current study compared to Stone et al.’s (2008) study may be due to qualitative differences in the sample (at-risk children from general population vs. siblings of children with ASD). The higher PPV resulting from two-level screening compared to Level 1 screening is promising, as is the increased sensitivity when using the new STAT cutoffs. However, final data on the psychometric properties of the M-CHAT-R & FUI alone have not yet been published for comparison in determining the efficacy of multi-level screening. Efforts must continue to reduce the false positive rate without significantly increasing the number of missed cases. Empirical studies are also needed to inform policy decisions on the early detection of ASD to shorten the delay in receiving appropriate treatment.
See more of: Clinical Phenotype
See more of: Symptoms, Diagnosis & Phenotype